Kinds of AI

Useful distinctions

Narrow AI performs specific tasks (recognising faces, recommending films, translating text) and is what exists today, even when a single model handles many tasks. Artificial general intelligence (AGI) refers to hypothetical systems with broad human-level competence; when or whether it arrives is debated. Methods split roughly into symbolic AI (explicit knowledge and reasoning, transparent but brittle) and statistical / learning-based AI (patterns from data, flexible but harder to explain). Modern systems often combine them: a language model with tools, retrieval and search.

Approaches compared

Strengths and weaknesses.

approach         knowledge comes from     strengths                    weaknesses
search/planning  problem model            optimal plans, guarantees    needs a model, can be slow
logic/rules      experts write rules      transparent, auditable       brittle, costly to maintain
probabilistic    models + data            handles uncertainty          needs structure and data
machine learning examples                 flexible, scales with data   opaque, data-hungry, biased data
hybrid (LLM+tools) pretraining + tools     broad, adaptable             errors, cost, evaluation hard

Use the simplest method that works

A rule or a search algorithm is often more reliable and explainable than a learned model for well-defined problems.

Quick check: What kind of AI exists in deployed systems today?

• AI with no data or rules
• Proven artificial general intelligence
• Conscious machines
• Narrow AI focused on specific tasks